def test_skip_last_raise():
model, param_dist, X, y, rng = setup()
search = HyperbandSearchCV(model,
param_dist,
skip_last=10,
random_state=rng)
search.fit(X, y)
def test_min_resource_param():
model, param_dist, X, y, rng = setup()
search = HyperbandSearchCV(model,
param_dist,
min_iter=3,
random_state=rng,
verbose=1)
search.fit(X, y)
assert (search.cv_results_['param_n_estimators'].data.min() == 3)
def test_skip_last():
model, param_dist, X, y, rng = setup()
search = HyperbandSearchCV(model,
param_dist,
skip_last=1,
random_state=rng)
search.fit(X, y)
# 177 Because in every round the last search is dropped
# 187 - (1 + 1 + 1 + 2 + 5)
assert (len(search.cv_results_['hyperband_bracket']) == 177)
def train_hyperband(X_train, X_test, y_train, y_test, mtype, common_name_model,
problemtype, classes, default_featurenames,
transform_model, settings, model_session):
# install
curdir = os.getcwd()
os.chdir(prev_dir(os.getcwd()) + '/training/helpers/hyperband')
os.system('python3 setup.py install')
from hyperband import HyperbandSearchCV
os.chdir(curdir)
# training and testing sets
files = list()
model_name = common_name_model + '.pickle'
if mtype in ['classification', 'c']:
model = RandomForestClassifier()
param_dist = {
'max_depth': [3, None],
'max_features': sp_randint(1, 11),
'min_samples_split': sp_randint(2, 11),
'min_samples_leaf': sp_randint(1, 11),
'bootstrap': [True, False],
'criterion': ['gini', 'entropy']
}
search = HyperbandSearchCV(model,
param_dist,
resource_param='n_estimators',
scoring='roc_auc')
search.fit(X_train, y_train)
params = search.best_params_
print('-----')
print('best params: ')
print(params)
print('------')
accuracy = search.score(X_test, y_test)
# SAVE ML MODEL
modelfile = open(model_name, 'wb')
pickle.dump(search, modelfile)
modelfile.close()
elif mtype in ['regression', 'r']:
print('hyperband currently does not support regression modeling.')
model_name = ''
model_dir = os.getcwd()
files.append(model_name)
return model_name, model_dir, files
def test_multimetric_hyperband():
model, param_dist, X, y, rng = setup()
# multimetric scoring is only supported for 1-D classification
first_label = (y == 1)
y[first_label] = 1
y[~first_label] = 0
multimetric = ['roc_auc', 'accuracy']
search = HyperbandSearchCV(model,
param_dist,
refit='roc_auc',
scoring=multimetric,
random_state=rng)
search.fit(X, y)
assert ('mean_test_roc_auc' in search.cv_results_.keys())
assert ('mean_test_accuracy' in search.cv_results_.keys())
assert (len(search.cv_results_['hyperband_bracket']) == 187)
def test_check_resource_param():
model, param_dist = setup()
HyperbandSearchCV(model, param_dist,
resource_param='wrong_name')._validate_input()
def test_check_eta():
model, param_dist = setup()
HyperbandSearchCV(model, param_dist, eta=0)._validate_input()
def test_check_skip_last():
model, param_dist = setup()
HyperbandSearchCV(model, param_dist, skip_last=-1)._validate_input()
def test_check_min_iter_smaller_max_iter():
model, param_dist = setup()
HyperbandSearchCV(model, param_dist, min_iter=30,
max_iter=15)._validate_input()
def test_check_max_iter():
model, param_dist = setup()
HyperbandSearchCV(model, param_dist, max_iter=-1)._validate_input()
#'min_child_weight' : [],
#'min_child_samples' : [],
#'subsample' : [],
#'subsample_freq' : [],
#'colsample_bytree' : [],
#'reg_alpha' : [],
#'reg_lambda' : [],
'n_jobs': [-1]
}
# In[42]:
search = HyperbandSearchCV(hb_lgb_model,
lgb_hb_param_dict,
cv=3,
verbose=1,
max_iter=200,
min_iter=50,
scoring='neg_log_loss')
# In[43]:
search.fit(x_train, y_train)
# In[44]:
search.best_params_
# # Fit new
# In[56]:
from scipy.stats import randint as sp_randint
from sklearn.datasets import load_digits
from sklearn.ensemble import RandomForestClassifier
from sklearn.preprocessing import LabelBinarizer
if __name__ == '__main__':
model = RandomForestClassifier()
param_dist = {
'max_depth': [3, None],
'max_features': sp_randint(1, 11),
'min_samples_split': sp_randint(2, 11),
'min_samples_leaf': sp_randint(1, 11),
'bootstrap': [True, False],
'criterion': ['gini', 'entropy']
}
digits = load_digits()
X, y = digits.data, digits.target
y = LabelBinarizer().fit_transform(y)
search = HyperbandSearchCV(model,
param_dist,
resource_param='n_estimators',
scoring='roc_auc',
n_jobs=1,
verbose=1)
search.fit(X, y)
print(search.best_params_)
print(search.best_score_)